Defining the optimal temporal and spatial resolution for cardiovascular magnetic resonance imaging feature tracking

The study population consisted of 12 healthy subjects and 9 patients with known ischemic heart disease and symptoms of HF. The study was approved by the Ethics Committee of the Charité-University Medicine Berlin and complied with the Declaration of Helsinki. All individuals gave written informed consent before participating in the study. Prospective electrocardiogram (ECG)-gated bSSFP cine sequences were acquired on a clinical 1.5 T CMR scanner (Achieva, Philips Healthcare, Best, the Netherlands) for long-axis 2 chamber (2Ch)- and 4-chamber (4Ch) views as well as a short axis stack. Imaging parameters included temporal resolutions of 20, 30, 40 and 50 frames/cardiac cycle. The reconstructed cardiac phases were obtained using a fixed interpolation factor of 50%. Spatial resolutions comprised high (1.5 × 1.5 mm in plane and 5 mm through-plane), standard (1.8 × 1.8 mm in plane and 8 mm through-plane) as well as low (3.0 × 3.0 mm in plane and 10 mm through-plane) settings (Fig. 1). For the purposes of this manuscript, the resolutions are referred to by their through-plane value of 5 mm, 8 mm and 10 mm respectively.

CMR-FT was performed using the commercially available software QStrain (version 3.1.16.9, Medis Medical Imaging Systems, Leiden, Netherlands, Fig. 1 and Additional file 1: Figure S1) for left ventricular (LV) global as well as segmental longitudinal and circumferential strain as well as associated systolic strain rates (GLS/GCS GLS SR/ GCS SR) and right ventricular (RV) global longitudinal strains (RV GLS). LV and RV were tracked at endocardial borders. The LV was tracked in long axis (2Ch and 4Ch) and short axis orientations. The RV was tracked in the 4Ch only [21]. Contours were manually traced in end-diastole and end-systole, the tracking algorithm was then applied following the tissue features over the cardiac cycle. Strain is reported as the percentage value of absolute shortening (Strain [%] = \(\frac{Length\;\text{endsystole }-Length\;\text{enddiastole}}{Length\;\text{enddiastole}}\) * 100). Accuracy was visually reviewed, where appropriate corrections were made by the observer to the initial contours only. Short axis strains (GCS) were assessed at basal (requiring circular myocardium and the absence of LV outflow tract during the cardiac cycle), midventricular (with both papillary muscles visible) and apical (maintained blood-pool throughout the cardiac cycle) positions. Intra-observer and inter-observer reproducibility was calculated within two operators in a core-laboratory with proven excellent intra- and inter-observer reproducibility in previous trials [21, 22, 28]. Both observers were well experienced in CMR and received dedicated training in CMR. Observer one performed two rounds 4 weeks apart for the assessment of intra-observer reproducibility, observer two performed one round for the assessment of inter-observer reproducibility respectively. Each round consisted of 3 repeated measurements, final strain values were based on the average to enhance reproducibility and decrease variability [21, 22]. Reproducibility of GLS and GCS was assessed on segmental level based on the segments proposed in the American Heart Association 16 segments model [29].

Intra- and inter-observer variability was evaluated based on Bland–Altman analyses [mean difference between measurements and corresponding 95% confidence interval (CI)] [30], intra-class correlation coefficients (ICC) for absolute agreements and the coefficient of variation (CoV, SD of mean difference divided by the mean \(\frac{SD (MD)}{mean}\)) [20]. An ICC > 0.74 was considered excellent, good between 0.60 and 0.74, fair between 0.4 and 0.59 and poor below 0.4. Continuous variables are expressed as mean ± standard deviation (SD). Dependent variables were tested using the Wilcoxon signed-rank test. An alpha level of 0.05 was considered statistically significant. Statistical analyses were performed using SPSS (version 24 for Windows, Statistical Package for the Social Sciences, International Business Machines, Inc., Armonk, New York, USA) and Excel (Microsoft Corporation, Redmond, Washington, USA).

The study population comprised 12 healthy subjects and 9 HF patients. Healthy subjects had a median age of 29 years (IQR 25, 34) with normal LV ejection fraction (LVEF) (median 60%; IQR 59, 60). HF patients had a median age of 67 years (ICR 51, 69) with LVEF (median 56%; IQR 47, 57). The median heart rate during CMR acquisition was 66/min (IQR 60, 71). LV strain and SR as well as RV strain values are reported in association to spatial and temporal resolution in Table 1.

Mean differences as well as corresponding SD, ICC and CoV of assessed strain and SR values are reported in Figs. 3 and 4 as well as in the supplements for LV GLS (Table 2), LV GLS SR (Table 3), LV GCS (Table 4), LV GCS SR (Table 5) and RV GLS (Additional file 1: Table S1) and Additional file 1: Figures S2–S13. Reproducibility on segmental level was excellent for LV GLS and GCS (ICC ≥ 0.81) as well as associated strain rates (ICC ≥ 0.77). Reproducibility in segments with wall motion abnormalities was good to excellent for GLS and GCS (ICC ≥ 0.73) and associated strain rates (ICC ≥ 0.82) (Additional file 1: Tables S2–S5).

Considering normal LV GLS values of − 20% with a 95% CI of − 19.3 to − 20.9 [36], a deviation of 1.8% comparing the lowest and the highest temporal resolution is close to 10% in relative change of absolute strain values. This could be considered clinically relevant. Nevertheless, variability caused by temporal resolution attenuated beyond 30 frames/cardiac cycle, which is in line with data coming from STE [27]. This underlines that consistent strain assessment is achieved at 30 frames/cardiac cycle for clinical imaging. Importantly, LV strain evaluation has recently been included within the guidelines for cardiac functional assessment in case of preserved LVEF such as in HFpEF [8]. Strain rate imaging expands comprehensive cardiac functional evaluations with the possibility of velocity assessments as established in echocardiography [8]. For strain rate values a continuous increase of absolute values assessed by FT is observed with increasing temporal resolution. Hence, if one aims to image velocities or acceleration, temporal resolution gains further importance and a threshold of 30 frames/cardiac cycle may not be sufficient for reliable assessment, especially when it comes to ECG-gated reconstructed CMR images. To date, introduction of clinical deformation imaging has not been fully achieved since the different available techniques are not completely interchangeable [31] including deviations introduced by different vendors within a given approach [22, 28]. However, data also indicates high correlation of CMR-derived deformation imaging, pointing towards valid measurements by all different approaches, which however require specific reference values [37]. Lastly, in opposition to previous studies [20], the current paper now with the evolution of the underlying software algorithms indicates excellent CMR-FT reproducibility also on the segmental level, which makes this technology applicable to a larger disease spectrum including regional myocardial disease such as coronary disease. However, further harmonisation between vendors and/or specific reference values are required to allow full clinical implementation of this promising technology [11, 13].